On Fri, Feb 21, 2020 at 5:09 PM Robin Murphy <robin.murphy@xxxxxxx> wrote: > On 21/02/2020 1:42 pm, Ard Biesheuvel wrote: > > OK, so where should it be instead? Are you saying it is now required > > to have a /soc or /smb node that contains all the device nodes? > > That is now my understanding, yes - if there is some property of the > system interconnect which is not entirely transparent relative to the > conceptual CPU-visible physical address space and needs to be described, > then the only way to correctly encode that in DT is with an explicit > level of 'bus' in the hierarchy to represent that interconnect. > Otherwise these address translation properties quickly end up in > philosophical conundrums over what the parent of the root node is. This confirms my similar understanding. Rob sent a patch to remove the root level dma-ranges from the ARM Integrator, and I have since added it back below the logic module bus after you pointed out that the dma-ranges was actually there for that very bus. Yours, Linus Walleij
